issai/Kazakh_Speech_Commands_Dataset

Kazakh Speech Commands Dataset

This dataset contains Kazakh speech commands generated using a combination of synthetic data generation and speech corpus scraping techniques. The synthetic data was created using the Piper neural text-to-speech system, leveraging five available Kazakh voices. The scraped data was extracted from a large-scale speech corpus using the Vosk Speech Recognition Toolkit. Further data augmentation techniques were applied to expand the dataset size. The dataset is accompanied by Jupyter notebooks detailing the data generation, scraping, and augmentation processes, as well as a directory containing information on model training, validation, and testing using a Keyword-MLP approach. Video tutorials for each step are available on YouTube.

Dataset Structure

The dataset is organized into the following components:

• Synthetic Data Generation: Jupyter Notebook (synthetic_data_generation.ipynb) demonstrating the process of generating synthetic Kazakh speech commands using Piper. The required Piper model can be downloaded from a provided Google Drive link.
• Speech Corpus Scraping: Jupyter Notebook (speech_corpus_scraping.ipynb) illustrating the use of the Vosk Speech Recognition Toolkit to extract speech commands from a large-scale corpus.
• Data Augmentation: Jupyter Notebook (data_augmentation.ipynb) detailing the audio augmentation techniques used to expand the dataset.
• Model Training, Validation, and Testing: The Keyword-MLP directory contains information related to model training, validation, and testing.
• Video Tutorials: A YouTube playlist provides video tutorials for each project step.

Dataset Creation

The dataset creation process involved three main stages:

1. Synthetic Data Generation: Leveraging the Piper neural text-to-speech system to create synthetic Kazakh speech commands.
2. Speech Corpus Scraping: Employing the Vosk Speech Recognition Toolkit to automatically extract relevant speech commands from a large-scale speech corpus.
3. Data Augmentation: Applying audio augmentation methods to increase the size and diversity of the dataset.

Usage

The provided Jupyter notebooks offer clear examples for replicating the dataset creation process. The Keyword-MLP directory contains information on model training, allowing researchers to build upon this dataset for speech command recognition tasks in the Kazakh language.

Links

• Piper (Text-to-Speech System): https://github.com/rhasspy/piper
• Vosk Speech Recognition Toolkit: https://github.com/alphacep/vosk-api/tree/master
• YouTube Tutorials: https://www.youtube.com/playlist?list=PLYwixe_5vr_kpH8_iLBSs4hWDN-50GgGo
• Project Repository: https://github.com/IS2AI/Kazakh-Speech-Commands-Dataset